How much VRAM does GLM-5.3 need?
GLM-5.3 needs 425 GB for its weights at 4-bit, 427 GB with an 8K context and 429 GB with 32K.
- Parameters
- 753B
- Active per token
- 40B
- Longest context
- 1M
- Released
- Aug 2026
- License
- GLM-5.3 License
- Strongest at
- Chat, Coding, Agents, Docs
Memory at each precision
| Precision | Weights | With 8K | With 32K | With 128K |
|---|---|---|---|---|
| As released (FP8) | 704 GB | 705 GB | 708 GB | 717 GB |
| 6-bit | 575 GB | 577 GB | 579 GB | 588 GB |
| 5-bit | 499 GB | 501 GB | 503 GB | 512 GB |
| 4-bit | 425 GB | 427 GB | 429 GB | 438 GB |
| 3-bit | 343 GB | 345 GB | 347 GB | 356 GB |
| 2-bit | 263 GB | 265 GB | 267 GB | 276 GB |
| 1.6-bit | 145 GB | 146 GB | 149 GB | 158 GB |
One conversation, 16-bit context cache, runtime overhead included.
What runs it
| Machine | Best fit | Context | Speed | First reply |
|---|---|---|---|---|
| RTX 3060 12 GB | Too big | |||
| RTX 4060 Ti 16 GB | Too big | |||
| RTX 3090 | Too big | |||
| RTX 4090 | Too big | |||
| RTX 5090 | Too big | |||
| 2× RTX 3090 | Too big | |||
| RTX PRO 6000 Blackwell | Too big | |||
| M4 Max 128 GB | Too big | |||
| M5 Max 128 GB | Too big | |||
| Ryzen AI Max+ 395 128 GB | Too big | |||
| DGX Spark 128 GB | Too big | |||
| M3 Ultra 256 GB | 1.6-bit | 384K | 38 tok/s | 25 s |
| M5 Ultra 512 GB | 3-bit | 216K | 28 tok/s | 7.8 s |
| H100 SXM | Too big | |||
| DGX Station GB300 | 1.6-bit | 960K | 158 tok/s | 0.7 s |
| 8× H200 NVL | FP8 | 1M | 60 tok/s | 0.3 s |
Best fit is the highest quality, up to 8-bit, that leaves 16K of context. First reply is the wait for a 2K prompt. Speeds are rough, plus or minus 25%.
How to run it
- llama.cpp
llama-server -hf unsloth/GLM-5.3-GGUF- vLLM
vllm serve zai-org/GLM-5.3- MLX (Mac)
mlx_lm.chat --model mlx-community/GLM-5.3-4bit
Questions
How much VRAM does GLM-5.3 need?
GLM-5.3 needs 425 GB for its weights at 4-bit, 427 GB with an 8K context and 429 GB with 32K.
Can GLM-5.3 run on an RTX 4090?
Not on one card. The smallest machine here that runs it is the M3 Ultra 256 GB.